How ibrahimethemhamamci/dentex works
This repository packages the DENTEX challenge evaluation entry point and two Dockerized HierarchicalDet algorithm phases. The main algorithm workflow configures and trains a hierarchical dental detector, then exposes prediction and evaluation; the intended result is abnormal-tooth boxes with quadrant, tooth-enumeration, and diagnosis labels. The README documents the three annotation levels and evaluation on fully labeled data. Source excerpts confirm detector/head/loss and COCO-style evaluation responsibilities; several training and inference connections are only indicated by the source index, so those links are marked unverified.
Main entry points: detector.py Diffusion detector · process.py Training entry point · predictor.py Prediction API · coco_3class_eval.py Hierarchical metrics
Components 16
Algorithm workflow
- Training entry point
HierarchialDet-FinalPhase-Docker/process.py - Prediction API
HierarchialDet-FinalPhase-Docker/hierarchialdet/predictor.py - Dental datasets
HierarchialDet-FinalPhase-Docker/detectron2/data/datasets/coco.py - Dataset mapping
HierarchialDet-FinalPhase-Docker/hierarchialdet/dataset_mapper.py - Data loading
HierarchialDet-FinalPhase-Docker/hierarchialdet/loader.py - Dental findings
Detection model
- Diffusion detector
HierarchialDet-FinalPhase-Docker/hierarchialdet/detector.py - Detection head
HierarchialDet-FinalPhase-Docker/hierarchialdet/head.py - Training losses
HierarchialDet-FinalPhase-Docker/hierarchialdet/loss.py - Swin backbone
HierarchialDet-FinalPhase-Docker/hierarchialdet/swintransformer.py
Detectron2 services
- Model configuration
HierarchialDet-FinalPhase-Docker/detectron2/config/config.py - Boxes and instances
HierarchialDet-FinalPhase-Docker/detectron2/structures/boxes.py - Training loop
HierarchialDet-FinalPhase-Docker/detectron2/engine/train_loop.py
Challenge evaluation
- Hierarchical metrics
HierarchialDet-FinalPhase-Docker/hierarchialdet/util/coco_3class_eval.py - Challenge evaluator
DentexChallenge-AlgorithmPhases-Evaluation-Docker/evaluation.py
Other components
- Algorithm participant
Connections 19
- Algorithm participant → Training entry point: starts training
- Training entry point → Model configuration: loads settings
- Training entry point → Training loop: runs training
- Training entry point → Prediction API: imports predictor
HierarchialDet-FinalPhase-Docker/process.py - Training entry point → Dental datasets: configures datasets
- Dental datasets → Dataset mapping: supplies records
- Dataset mapping → Data loading: maps samples
- Data loading → Training loop: feeds batches
- Training loop → Diffusion detector: invokes model
- Diffusion detector → Detection head: calls head
HierarchialDet-FinalPhase-Docker/hierarchialdet/detector.py - Diffusion detector → Training losses: computes loss
HierarchialDet-FinalPhase-Docker/hierarchialdet/detector.py - Diffusion detector → Swin backbone: builds backbone
- Diffusion detector → Boxes and instances: uses instances
HierarchialDet-FinalPhase-Docker/hierarchialdet/detector.py - Prediction API → Diffusion detector: runs inference
- Prediction API → Dental findings: returns predictions
- Algorithm participant → Challenge evaluator: submits results
- Challenge evaluator → Hierarchical metrics: scores predictions
- Hierarchical metrics → Dental datasets: reads annotations
HierarchialDet-FinalPhase-Docker/hierarchialdet/util/coco_3class_eval.py - Hierarchical metrics → Dental findings: reports metrics